Comparison of nearest neighbour and neural network based classifications of patient's activity

This paper presents a comparison of 1-nearest neighbour (1-NN) and neural network based classification of patient activity. The data for classification was acquired from two 6 degree-of-freedom accelerometers deployed at the wrists of a patient. Instead of calculating statistical values, we studied the use of data samples acquired from 200ms time window. The best results were achieved with the 1-nearest neighbour algorithm. The overall accuracy of the 1-NN method was nearly 100%. The learning method for neural network used was the backpropagation with momentum. According to our experiments, the results of classification were more accurate with 1-NN in comparison with the result of neural network (93.4%).